TrebuchetDynamics · XelHaku · May 11, 2026 · May 8, 2026 · May 8, 2026 · May 8, 2026
diff --git a/README.md b/README.md
@@ -174,7 +174,7 @@ CI runs `go test ./... -count=1`, `go run ./cmd/progress validate`, and `git dif
 | Phase | Status | Shipped |
 |-------|--------|---------|
 | Phase 1 — The Dashboard | ✅ | 4/4 subphases |
-| Phase 2 — The Gateway | ✅ | 21/21 subphases |
+| Phase 2 — The Gateway | 🔨 | 20/21 subphases |
 | Phase 3 — The Black Box (Memory) | ✅ | 15/15 subphases |
 | Phase 4 — The Brain Transplant | ✅ | 13/13 subphases |
 | Phase 5 — The Final Purge | 🔨 | 18/23 subphases |

diff --git a/internal/gateway/render_test.go b/internal/gateway/render_test.go
@@ -130,7 +130,7 @@ func TestFormatToolProgressPlain_ToolTraceFixtureMatrix(t *testing.T) {
 	}
 }
 
-func TestFormatToolProgressPlain_TruncatedWebPreviewsKeepRightEdge(t *testing.T) {
+func TestFormatToolProgressPlain_TruncatedWebPreviewsKeepLeftEdge(t *testing.T) {
 	f := kernel.RenderFrame{SoulEvents: []kernel.SoulEntry{
 		{At: time.Now(), Text: "tool: web_extract: https://docs.openclaw.ai/concepts/multi-agent"},
 		{At: time.Now(), Text: "tool: browser_navigate: https://docs.openclaw.ai/concepts/multi-agent"},
@@ -139,12 +139,12 @@ func TestFormatToolProgressPlain_TruncatedWebPreviewsKeepRightEdge(t *testing.T)
 
 	got := FormatToolProgressPlain(f)
 	for _, want := range []string{
-		`📄 web_extract: "...docs.openclaw.ai/concepts/multi-agent"`,
-		`🌐 browser_navigate: "...docs.openclaw.ai/concepts/multi-agent"`,
-		`🔍 web_search: "...concepts/multi-agent strings in tools"`,
+		`📄 web_extract: "https://docs.openclaw.ai/concepts/mul..."`,
+		`🌐 browser_navigate: "https://docs.openclaw.ai/concepts/mul..."`,
+		`🔍 web_search: "site:docs.openclaw.ai/concepts/multi-..."`,
 	} {
 		if !strings.Contains(got, want) {
-			t.Fatalf("FormatToolProgressPlain missing right-edge preview %q in:\n%s", want, got)
+			t.Fatalf("FormatToolProgressPlain missing left-edge preview %q in:\n%s", want, got)
 		}
 	}
 }
@@ -191,8 +191,8 @@ func TestFormatToolProgressPlain_MineruGatewayTranscriptShape(t *testing.T) {
 	for _, want := range []string{
 		`📚 skill_view: "gormes-hermes-parity"`,
 		`⏰ cronjob: "list"`,
-		`🌐 browser_navigate: "...www.reddit.com/r/WebAfterAI/s/example"`,
-		`🌐 browser_navigate: "...old.reddit.com/r/WebAfterAI/s/example"`,
+		`🌐 browser_navigate: "https://www.reddit.com/r/WebAfterAI/s..."`,
+		`🌐 browser_navigate: "https://old.reddit.com/r/WebAfterAI/s..."`,
 		`💻 terminal: "curl -L https://example.test/post.json"`,
 		`📸 browser_snapshot...`,
 		`💻 terminal: "curl -L https://example.test/post.json" (×2)`,

diff --git a/internal/progress/progress_test.go b/internal/progress/progress_test.go
@@ -459,8 +459,12 @@ func TestLoad_RealFile_Phase2ExecutionQueue(t *testing.T) {
 	if routing.Priority != "P1" {
 		t.Fatalf("Phase 2.B.5 priority = %q, want P1", routing.Priority)
 	}
-	if got := routing.DerivedStatus(); got != StatusComplete {
-		t.Fatalf("Phase 2.B.5 = %q, want complete — all session-context and delivery-routing rows shipped", got)
+	switch got := routing.DerivedStatus(); got {
+	case StatusComplete, StatusInProgress:
+		// complete or reopened by a newly discovered gateway slash-handler
+		// gap such as `Gateway /model interactive provider/model picker`
+	default:
+		t.Fatalf("Phase 2.B.5 = %q, want complete or in_progress", got)
 	}
 	routingItems := itemsByName(routing.Items)
 	topicCloseout := routingItems["Telegram topic mode off/help/auth/debounce closeout"]

diff --git a/internal/tooltrace/tooltrace.go b/internal/tooltrace/tooltrace.go
@@ -196,8 +196,13 @@ func quoteAndTruncate(toolName, s string, limit int) string {
 }
 
 func rightEdgePreviewTool(name string) bool {
+	// File-path arguments keep the tail (filename matters more than the
+	// directory chain). URLs and search queries keep the head: the domain
+	// or the first words of the query are the user-meaningful part, and a
+	// `"...et.com/market-data/..."` preview hides which site Gormes is
+	// actually visiting.
 	switch strings.TrimSpace(name) {
-	case "web_search", "web_extract", "web_crawl", "browser_navigate", "read_file", "write_file", "patch":
+	case "read_file", "write_file", "patch":
 		return true
 	default:
 		return false

diff --git a/internal/tooltrace/tooltrace_preview_test.go b/internal/tooltrace/tooltrace_preview_test.go
@@ -0,0 +1,55 @@
+package tooltrace
+
+import "testing"
+
+// Live regression 2026-05-10: an operator on Telegram saw tool-call previews
+// rendered as `🔍 web_search: "...et docs websocket WSS Quickstart CLOB"`
+// and `📄 web_extract: "...et.com/market-data/websocket/overview"`. Both
+// hide the user-meaningful start (the search query's first words and the
+// URL's domain) behind a leading "...". Search queries and URLs must keep
+// their head; only file-path tool args (read_file/write_file/patch) keep
+// the tail because the filename usually matters more than the directory
+// chain.
+
+func TestFormatPlain_PreservesUrlAndQueryHead(t *testing.T) {
+	cases := []struct {
+		name string
+		in   string
+		want string
+	}{
+		{
+			name: "web_search keeps head of query",
+			in:   "tool: web_search: polymarket docs websocket WSS Quickstart CLOB",
+			want: `🔍 web_search: "polymarket docs websocket WSS Quickst..."`,
+		},
+		{
+			name: "web_extract keeps domain at head of URL",
+			in:   "tool: web_extract: https://docs.polymarket.com/market-data/websocket/overview",
+			want: `📄 web_extract: "https://docs.polymarket.com/market-da..."`,
+		},
+		{
+			name: "browser_navigate keeps domain at head of URL",
+			in:   "tool: browser_navigate: https://docs.polymarket.com/market-data/websocket/overview",
+			want: `🌐 browser_navigate: "https://docs.polymarket.com/market-da..."`,
+		},
+		{
+			name: "read_file keeps tail (filename)",
+			in:   "tool: read_file: internal/channels/telegram/document_cache.go",
+			want: `📖 read_file: "...l/channels/telegram/document_cache.go"`,
+		},
+		{
+			name: "write_file keeps tail (filename)",
+			in:   "tool: write_file: internal/channels/telegram/document_cache.go",
+			want: `🔧 write_file: "...l/channels/telegram/document_cache.go"`,
+		},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			got := FormatPlain(tc.in)
+			if got != tc.want {
+				t.Fatalf("FormatPlain(%q)\n got: %s\nwant: %s", tc.in, got, tc.want)
+			}
+		})
+	}
+}
diff --git a/scripts/gormes-builder-cron.sh b/scripts/gormes-builder-cron.sh
@@ -98,6 +98,85 @@ sync_development_branch() {
   die "$BRANCH diverged from origin/$BRANCH; skipping automated builder run"
 }
 
+opencode_prompt() {
+  cat <<'PROMPT'
+You are an autonomous coding agent (opencode + a local ollama model — typically
+qwen3-coder:30b MoE) running from cron inside the Gormes repository. The
+codexu/codex backend is rate-limited until 2026-05-14, so this loop is keeping
+forward motion with a smaller local model. Bias toward small, safe, finished
+work over big plans.
+
+Skill loading
+- The repo's gormes-* skills are SKILL.md instruction files, not MCP servers.
+- Use the `skill` tool to load a skill by name (e.g. `skill gormes-git`). The
+  tool returns the SKILL.md content; read it and follow the instructions
+  manually. Do NOT call `skill_mcp` for these — they have no MCP endpoint.
+- The full skill catalog is under docs/development-skills/gormes-*/SKILL.md.
+  The same files are symlinked into ~/.config/opencode/skills/ for
+  discoverability.
+
+Tool-schema reminders (opencode is strict about these — wrong shape = no-op)
+- `bash`: REQUIRES both `command` (the shell command) and `description` (one
+  short sentence describing what it does). Omitting `description` errors out.
+- `read`: takes `file_path` (absolute path preferred).
+- `edit`: takes `file_path`, `old_string`, `new_string`. The old_string must
+  match exactly (whitespace + newlines included) and be unique in the file.
+- `write`: takes `file_path` and `content`. Use only for new files; prefer
+  `edit` for changes to existing files.
+- When in doubt, run a single `bash` with `git status --short` first — it is
+  the cheapest way to ground yourself in the worktree state.
+
+Mandatory branch + safety rules
+- Stay on the existing `development` branch only. Do not create branches or
+  worktrees. If HEAD is not `development`, stop and report the blocker.
+- Preserve any user or parallel-agent changes you find. If you do not
+  understand untracked files (IDENTITY.md, SOUL.md, TOOLS.md, memory/*, etc.),
+  leave them alone — they are not your work.
+- Never force-push. Never bypass hooks. Never recreate cmd/planner-loop or
+  cmd/builder-loop.
+
+Single focused goal per cycle (pick the FIRST that applies)
+
+1. Worktree is dirty with YOUR previous-cycle work?
+   Load `gormes-git`, follow it to commit + push only the files you authored
+   this cycle (skip unfamiliar untracked files). Then exit cleanly.
+
+2. Worktree is dirty with files you cannot attribute to yourself?
+   Do not touch them. Run `go test ./... -count=1` and report the result.
+   Exit cleanly. Leave the worktree as you found it.
+
+3. Worktree is clean and you can read progress.json?
+   Pick ONE small, builder-ready row from
+   docs/content/building-gormes/architecture_plan/progress.json (priority P1
+   or P2, slice_size small, contract_status ready, blocked_by empty). Load
+   `gormes-builder` and `gormes-tdd-slice`, follow them to implement that
+   one row with a red→green→refactor loop. Run `go run ./cmd/progress
+   validate` and `git diff --check` afterwards. If validation fails, fix
+   the validation only (do not chase scope creep).
+
+4. You are unsure or any tool returns confusing output?
+   Stop. Write a one-line summary of what you tried and what blocked you.
+   Exit with status 0 — the next cron tick is in 60s, the user can read
+   the log.
+
+Hard constraints
+- Do exactly ONE bounded thing this cycle. Do not chain row-implementation +
+  parity sweep + planner work.
+- Do NOT add new progress.json rows. Do NOT do parity audits. Those are for
+  the codex backend which has the cognitive headroom.
+- If a step requires more than ~5 tool calls to complete, you are over your
+  size budget. Save what you have and exit.
+- Run validation gates immediately after edits, not at the end of a long
+  chain — partial green is better than untested wholeness.
+- Do not edit files outside the row's `write_scope`. If the row's scope is
+  unclear, fall back to goal #4 (report and exit).
+
+When you cannot continue
+- Use plain text "I am stopping because: <reason>". Do not pretend success.
+- Exit. The cron wrapper will retry in 60 seconds.
+PROMPT
+}
+
 codex_prompt() {
   cat <<'PROMPT'
 You are Codex running from cron inside the Gormes repository.
@@ -171,15 +250,19 @@ main() {
 
   if [[ "${1:-}" == "--dry-run" ]]; then
     log "dry run: would execute $BACKEND builder prompt with timeout $RUN_TIMEOUT"
-    codex_prompt
+    if [[ "$BACKEND" == "opencode" ]]; then
+      opencode_prompt
+    else
+      codex_prompt
+    fi
     exit 0
   fi
 
   set +e
   local status
   if [[ "$BACKEND" == "opencode" ]]; then
     local opencode_log="$LOG_DIR/$RUN_ID.opencode.jsonl"
-    codex_prompt | timeout "$RUN_TIMEOUT" "$resolved_bin" run \
+    opencode_prompt | timeout "$RUN_TIMEOUT" "$resolved_bin" run \
       --model "$OPENCODE_MODEL" \
       --dir "$REPO_ROOT" \
       --format json \

diff --git a/webpages/docs/content/building-gormes/architecture_plan/_index.md b/webpages/docs/content/building-gormes/architecture_plan/_index.md
@@ -87,12 +87,12 @@ classification and row-backed implementation queue.
 ## Progress
 
 <!-- PROGRESS:START kind=docs-full-checklist -->
-**Overall:** 87/100 subphases shipped · 10 in progress · 3 planned
+**Overall:** 86/100 subphases shipped · 11 in progress · 3 planned
 
 | Phase | Status | Shipped |
 |-------|--------|---------|
 | Phase 1 — The Dashboard | ✅ | 4/4 subphases |
-| Phase 2 — The Gateway | ✅ | 21/21 subphases |
+| Phase 2 — The Gateway | 🔨 | 20/21 subphases |
 | Phase 3 — The Black Box (Memory) | ✅ | 15/15 subphases |
 | Phase 4 — The Brain Transplant | ✅ | 13/13 subphases |
 | Phase 5 — The Final Purge | 🔨 | 18/23 subphases |
@@ -136,7 +136,7 @@ classification and row-backed implementation queue.
 - [x] Skill-pack coverage audit for Hermes-in-Go completion
 - [x] Canonical development-skills directory and loader symlinks
 
-## Phase 2 — The Gateway ✅
+## Phase 2 — The Gateway 🔨
 
 *Go-native operator wiring harness: tools, Telegram, shared gateway chassis, shipped cron, and the first OS-AI spine slices before focused channel closeout*
 
@@ -180,7 +180,7 @@ classification and row-backed implementation queue.
 - [x] WhatsApp outbound pairing gate + raw peer mapping
 - [x] WhatsApp reconnect backoff + send retry policy
 
-### 2.B.5 — Session Context + Delivery Routing ✅
+### 2.B.5 — Session Context + Delivery Routing 🔨
 
 - [x] Gateway session store + SessionSource parity
 - [x] Gateway manual reset session-boundary hooks
@@ -261,6 +261,7 @@ classification and row-backed implementation queue.
 - [x] Telegram streaming edit Markdown safety
 - [x] Telegram guest mention allowlist bypass
 - [x] Gateway platform reconnect isolation + channel health limits
+- [ ] Gateway /model interactive provider/model picker
 
 ### 2.B.10 — WeChat Adapter ✅